CADSynth: A Dataset for Machining Feature Recognition in B-rep Models

Machining Feature Recognition (MFR) is the process of grouping geometric and topological elements from CAD models from the perspective of manufacturing requirements to obtain machining features with specific engineering semantics, such as holes, slots, steps and fillets. This dataset contains 100,000 synthetic CAD models with machining feature labels, supporting data-driven deep learning methods for machining feature recognition.The CADSynth dataset provides three types of data files for each CAD model. These include a STEP file, which records the three-dimensional shape of the CAD model (compliant with ISO 10303-21 and GB/T 16656 standards), a JSON file that records the machining feature labels corresponding to each topological face (compliant with ECMA-404 and ISO/IEC 21778:2017 standards), and a B-rep data graph representation file (with a .bin extension), which can be used for deep learning training (formatted for the Deep Graph Library, an open-source graph neural network framework).

加工特征识别（Machining Feature Recognition，MFR）是指从制造需求的视角出发，对计算机辅助设计（Computer Aided Design，CAD）模型中的几何与拓扑元素进行分组，以获取具备特定工程语义的加工特征（如孔、槽、台阶与圆角）的过程。本数据集包含10万个带有加工特征标注的合成CAD模型，可支撑面向加工特征识别的数据驱动深度学习方法的研究与应用。CADSynth数据集为每个CAD模型提供三类数据文件：一是符合ISO 10303-21与GB/T 16656标准的STEP文件，用于记录CAD模型的三维形态；二是符合ECMA-404及ISO/IEC 21778:2017标准的JSON文件，用于记录每个拓扑面对应的加工特征标注；三是扩展名为.bin的边界表示（Boundary Representation，B-rep）数据图表征文件，该文件可用于深度学习训练，其格式适配开源图神经网络框架深度图库（Deep Graph Library，DGL）。

搜集汇总

背景与挑战

背景概述

CADSynth是一个专门用于B-rep模型加工特征识别的大规模数据集，包含100,000个合成CAD模型，每个模型提供STEP、JSON和B-rep图表示三种数据文件，支持深度学习方法的训练和应用。该数据集旨在通过标注加工特征（如孔、槽等），推动数据驱动的制造语义识别研究。